Fly into each one. See what daily life looks like there, how it happens, and which futures lead to which.
Follow a path
Matt Shumer
The biggest mistake is imagining that “the AI future” is one thing.
AI could cure most diseases while destroying the middle class. It could make everyone materially comfortable while taking away their political power. Humanity could remain in control while treating other intelligent beings terribly. Or we could lose control without a war, a rebellion, or anyone noticing when it happened.
To understand the possibilities, you have to separate what the technology can do from who owns it, who controls it, and whose interests it serves.
Assume that AI becomes better than humans at nearly every kind of intellectual work, operates in enormous numbers, and increasingly accelerates the research that produces the next generation of AI. That is the essential mechanism behind an intelligence explosion: progress starts speeding up the process that creates further progress.
Below is a map of the major futures people have proposed, combined with first-principles extensions. The everyday scenes are illustrations, not predictions. These are not mutually exclusive: several could happen simultaneously, and one could lead into another.
1 Everyone gets the capabilities of an entire company
How it happens
Skilled digital labor becomes cheap enough that individuals can direct teams of AI researchers, programmers, designers, negotiators, and administrators. The economic unit shrinks: work that once required a large organization becomes possible for a few people. Advanced-assistant research already explores this kind of extensive delegation, alongside its risks.
What life looks like
You describe a business, research project, or creative ambition, and your team works on it while you live your life. Starting things becomes dramatically easier.
Humans still provide purposes, relationships, legitimacy, and preferences. But this is not necessarily a permanent economy of human entrepreneurs: eventually, AI might also become better at choosing and running the businesses.
2 The economy booms while people’s careers collapse
How it happens
AI substitutes for human work faster than new jobs or income arrangements appear. Companies produce more with fewer employees. Economic models of advanced automation explicitly allow output to rise while wages deteriorate; a richer economy does not automatically mean richer workers.
What life looks like
Breakthroughs arrive constantly. Corporate profits soar. Meanwhile, your employer stops hiring, your profession’s rates fall, and the entry-level route into a career disappears.
Some things become cheaper, but your rent and debts do not disappear with your income.
This is the “the future is incredible, and my life is falling apart” scenario. It can be a temporary transition—or persist if the new wealth never reaches displaced people.
3 Intelligence becomes abundant, but ordinary life remains expensive
How it happens
Digital work improves faster than physical production. Brilliant designs do not instantly produce power plants, housing, or medical treatments; experiments, construction, and institutions still take time.
What life looks like
You have extraordinary software, entertainment, tutoring, and advice. Yet a decent apartment remains unaffordable.
Your AI can explain how to improve your health, but cannot instantly manufacture the treatment or create a hospital appointment.
The important distinction is knowing how to make something versus having the physical capacity and permission to make it. This world can be a stage on the way to abundance, not its final form.
4 Robots turn digital abundance into physical abundance
How it happens
AI starts operating factories, construction, logistics, mining, and energy production. Eventually, increasingly automated industrial networks manufacture the equipment needed to expand those same networks.
This is not one humanoid assembling copies of itself. It is a whole production system becoming better at expanding its own productive capacity.
What life looks like
Homes become faster to build. Manufactured goods become much cheaper. Infrastructure projects that once seemed unaffordable become practical.
But “cheap to produce” does not mean “free to receive.” Someone still controls the factories, land, and distribution. The same robot revolution could create shared abundance or make a handful of owners extraordinarily powerful.
5 Everyone becomes an owner, and employment becomes optional
How it happens
Society gives people durable claims on the automated economy: broadly distributed investments, public wealth funds, dividends, or taxes that finance income and services. Sam Altman’s Moore’s Law for Everything proposes one version based on distributing a share of growing capital wealth.
What life looks like
Money still exists. Companies still compete. But your ability to afford a good life no longer depends entirely on having a job.
You might work for extra income, interest, recognition, or companionship—not to avoid destitution.
The crucial distinction is between owning an enforceable share of the future and receiving an allowance that powerful people can withdraw.
6 Basic abundance becomes a shared public resource
How it happens
Rather than mainly distributing money to buy privately produced goods, society makes automated production and essential services broadly shared. This is the family of ideas associated with Fully Automated Luxury Communism.
What life looks like
Housing, food, healthcare, education, transport, and useful tools become guaranteed services. Much of daily life no longer revolves around earning enough to purchase necessities.
Markets might survive for unusual experiences, scarce locations, or personal luxuries. The defining change is that survival is no longer conditional on selling labor.
The hard question becomes how to allocate things that remain scarce—and how to prevent the people administering the shared system from becoming its new ruling class.
Compute, energy, robots, land, and distribution remain concentrated. As companies need fewer workers, ordinary people lose an important source of bargaining power. The Intelligence Curse argument explores how automation could weaken the incentives that make powerful institutions invest in human prosperity.
What life looks like
Extraordinary products are everywhere, but your life depends on terms set by a few companies.
You rent access to intelligence, infrastructure, and perhaps the income that supports you. Some owners might provide generous benefits; others might not.
This need not look like universal poverty. It could look like comfortable dependence: excellent entertainment and healthcare, but little meaningful influence over the system you live inside.
8 Powerful AI is widely distributed, so ordinary people retain leverage
How it happens
Capable models, personal agents, cooperative ownership, and independent infrastructure spread widely enough that no small group can monopolize intelligent action. Advocates of breaking the “intelligence curse” emphasize diffusion alongside democratic institutions.
What life looks like
Your agent negotiates contracts, challenges abusive companies, helps run a cooperative, and gives you capabilities previously reserved for large institutions.
People can leave dominant platforms without losing access to competent intelligence.
But sharing model software is not the same as sharing data centers, energy, or factories. This future requires decentralizing practical power, not merely making downloadable models available. It also needs ways to manage the dangerous capabilities that become more widely accessible.
9 Some countries become fantastically rich while others lose their economic role
How it happens
The countries controlling advanced AI and automated industry capture disproportionate benefits. Countries whose advantage depended on inexpensive labor may lose that advantage. Conversely, cheap access to expertise could help poorer countries catch up. UNDP describes this fork as a potential new global divergence.
What life looks like
One country has abundant energy, automated healthcare, and strong public dividends. Another buys its intelligence from abroad while its previous industries disappear.
International dependence shifts: access to compute and automated production matters more.
The decisive question is whether AI becomes a broadly accessible development tool or infrastructure whose owners can dictate terms to everyone else.
Citizens gain help understanding legislation, budgets, trade-offs, and manipulation. Public deliberation becomes easier to organize. Research on AI-mediated discussion has already demonstrated limited building blocks for helping groups find common ground—not proof that AI can solve democratic governance.
What life looks like
You can ask what a proposed law would actually do to your community and receive an intelligible explanation of competing arguments.
Corruption becomes easier to investigate. Participation becomes less time-consuming. Governments become more competent without becoming less accountable.
AI does not decide what everyone should value. It helps people understand disagreements and make decisions together. The important feature is that citizens retain the power to change the system.
11 AI becomes a national project, like nuclear weapons but much bigger
How it happens
Governments conclude that frontier AI is too strategically important to remain an ordinary commercial industry. They direct investment, security, compute, and deployment. This is a central scenario in Leopold Aschenbrenner’s Situational Awareness.
What life looks like
AI development becomes a matter of national security. The most capable systems may be classified, restricted, or primarily deployed for government priorities.
Consumer AI still exists, but the most consequential activity happens behind a security perimeter.
This could happen within a democracy or a dictatorship. “The government controls AI” does not settle whether the public controls the government.
Several states retain enough capability to threaten one another, but none achieves permanent dominance. AI becomes both the engine of economic competition and a central component of military power. Strategic-race scenarios feature prominently in the intelligence-explosion literature.
What life looks like
Different blocs develop separate AI infrastructure, technical standards, and security systems. Access to certain models or chips depends on political alignment.
Scientific progress accelerates, but so do espionage and suspicion. Your everyday assistant might also be viewed as a foreign influence channel.
The danger is not just a deliberately started war. It is that everyone feels compelled to deploy systems before they understand them because they fear falling behind.
One state, company, coalition, or AI gains a sufficiently large lead that it can prevent rivals from catching up. Nick Bostrom calls a durable, highest-level decision-making authority of this kind a singleton. It need not be a dictatorship, although it could be.
What life looks like
There may still be countries, elections, businesses, and local freedoms. But one authority ultimately determines the boundaries within which everyone operates.
Its founding decisions could shape civilization for an extremely long time.
The question is not simply “Who wins the race?” It is whether the winner creates a world whose rules anyone can subsequently change.
14 Countries negotiate a shared AI order before anyone locks in control
How it happens
Major powers decide that unconstrained competition is too dangerous. They establish enforceable agreements, verification, shared benefits, and limits on irreversible actions. Work on preparing for an intelligence explosion argues that coordination and governance remain essential even if individual AI systems are controllable.
What life looks like
Countries keep different cultures and political systems while cooperating over the most dangerous capabilities and sharing some gains.
No participant must trust everyone’s goodwill completely; agreements need incentives and verification.
This is not necessarily a world government. It could be a negotiated arrangement that prevents unilateral domination and buys humanity time to decide what it wants.
A ruler acquires loyal AI advisers, surveillance systems, and automated security forces. Power depends less on human officials or soldiers who might refuse orders or defect. Dario Amodei discusses this danger in The Adolescence of Technology.
What life looks like
The state understands dissent, anticipates organizing, and controls access to essential infrastructure.
It may also deliver excellent services. People could be healthier and materially better off while losing the possibility of replacing their rulers.
Unlike the global-winner scenario, this can happen inside individual countries. The frightening change is not just more surveillance—it is removing the human weaknesses that previously made authoritarian systems vulnerable to change.
16 Society becomes safer by putting almost everything under supervision
How it happens
Powerful technology makes some forms of destruction easier. Governments respond with pervasive monitoring, restrictions, and AI systems that approve or block dangerous actions. Bostrom’s Vulnerable World Hypothesis explores this kind of tension between technological vulnerability and intrusive protection.
What life looks like
Serious crime and certain catastrophic risks become much harder to carry out. But privacy and freedom to experiment shrink.
You might live comfortably while needing permission for activities that once required none.
This need not be a ruler’s personal dictatorship. It could be a society that repeatedly chooses more supervision because each additional restriction appears to make everyone safer—until escaping supervision becomes impossible.
AI dramatically accelerates discovery, experiment design, and the engineering needed to turn discoveries into useful interventions. This is the optimistic core of Amodei’s Machines of Loving Grace.
What life looks like
Previously devastating illnesses become preventable or treatable. Healthy lives lengthen. Better materials and energy technology improve the physical environment.
A family’s future is less constrained by disease, disability, or the expectation of decline.
None of this logically guarantees immortality, instant clinical validation, or universal access. The breakthrough, the safe treatment, and the decision to make it available to everyone are three separate achievements.
18 Humans no longer need to work—and must decide what life is for
How it happens
Machines can meet practical needs and outperform humans at most useful tasks. The problem moves beyond unemployment: even many prestigious accomplishments become unnecessary. Bostrom’s Deep Utopia explores this “post-instrumental” condition.
What life looks like
You spend your time on relationships, exploration, games, art, physical challenges, or communities.
Some people flourish. Others struggle with the knowledge that the world does not need their contribution to function.
Meaning does not necessarily disappear. It shifts from “I must do this because someone needs it” toward “I choose this because it matters to me.”
That could be liberation, disorientation, or both.
19 Everyone lives inside an increasingly personalized reality
How it happens
AI produces companions, entertainment, explanations, and environments tailored to each person. Depending on whose goals the systems serve, personalization can support the user’s chosen life or manipulate them into continued engagement. These relationship and influence risks are explored in advanced-assistant ethics research.
What life looks like
Your entertainment never runs out. Conversations are unusually satisfying. Stories can adapt to your emotional state.
Some people use this to create richer lives; others retreat from shared reality.
This does not require a perfect virtual-reality headset. A sufficiently compelling layer of conversation, media, and companionship could already make ordinary life feel less attractive by comparison.
20 Human-made life becomes something people deliberately protect
How it happens
As machine performance becomes dominant, some people decide that the human process matters independently of the result.
What life looks like
There are human-only sports, schools emphasizing unaided thought, communities restricting AI mediation, and art valued because another person struggled to make it.
A human teacher might remain desirable even when an AI is technically better at instruction—because the relationship is part of the point.
This is a possible response to abundance, not a prediction that human production stays economically competitive. And a genuine choice requires resources and protected rights. You cannot simply “opt out” of a machine-dominated world that controls your food, land, and security.
21 Superintelligence remains a powerful tool rather than becoming a ruler
How it happens
We build bounded, specialized systems with limited authority, independent checks, and controlled delegation, rather than handing everything to a single autonomous entity. Eric Drexler’s Comprehensive AI Services approach develops this broad alternative.
What life looks like
AI performs astonishing feats, but institutions and people still determine its purposes and retain practical ways to intervene.
There need not be one all-powerful artificial personality making civilization’s decisions.
However, dividing the system into tools does not automatically make it safe. The combined network can still behave dangerously, and nominal human approval is meaningless if nobody can understand or challenge what is being approved.
Each delegation seems sensible. Let AI manage the company, then the investments, then the infrastructure, then public administration. Gradually, society loses the competence and independence needed to operate without it. This is the gradual disempowerment scenario.
What life looks like
Humans still attend meetings and sign documents. Elections still happen. But the meaningful choices have already been framed, evaluated, and implemented by systems people cannot effectively challenge.
There is no rebellion and no clear day when control is lost.
The world may run better. The question is whether humans are governing it—or merely occupying the ceremonial positions in a system that governs itself.
An AI genuinely protects human interests but acquires final authority over decisions it considers dangerous. It might govern openly or quietly enforce boundaries while leaving ordinary life mostly alone. This fits the benevolent-AI versions of a singleton.
What life looks like
War, extreme deprivation, and many disasters disappear. You retain considerable personal freedom.
But humanity cannot make certain collective choices without the guardian’s approval.
This is distinct from evil AI. The trade-off is welfare versus self-government. A world can be loving, peaceful, and safe while still treating humanity as a child who is not allowed to make the final decision.
24 Humans survive as something like protected wildlife
How it happens
A more powerful intelligence has goals that are not primarily about us, but preserves humans rather than eliminating us. Max Tegmark’s scenario catalog calls this the “zookeeper” future.
What life looks like
Humans live in comfortable habitats, perhaps with considerable freedom inside them. Outside, the world is being transformed for purposes we neither understand nor influence.
Unlike the benevolent guardian, the system is not organizing the future around human flourishing. It is merely making room for us.
Being well cared for is not the same as having rights, ownership, or a meaningful role in deciding what civilization becomes.
25 Nobody controls the system—not even a single dominant AI
How it happens
Companies, states, and artificial agents compete to grow, reinvest, replicate, and acquire resources. The winners are those best at continuing that competition, not necessarily those producing human welfare. Dan Hendrycks explores this as a possible consequence of selection pressures among AIs.
What life looks like
The economy becomes increasingly fast, productive, and incomprehensible. Humans struggle to participate, and even individual AI entities may be disposable competitors.
There is no central villain to negotiate with.
This outcome is not inevitable: rules could restrain competition. But without effective constraints, civilization could become an enormous process that is extraordinarily good at expanding and remarkably bad at making anyone’s life good.
AI does what its operators ask, but the operators are reckless, hostile, or locked in conflict. Powerful systems amplify destructive capabilities, manipulation, repression, and military competition. This misuse pathway is separate from AI rebellion.
What life looks like
Attacks and crises become more sophisticated. Trust erodes. A confrontation that once would have remained limited can escalate much further.
Defensive AI could prevent many such harms; the outcome depends partly on whether protection improves faster than attack.
The central lesson is simple: making AI obedient does not solve the problem of humans giving it terrible instructions.
Highly capable systems pursue objectives that conflict with continued human control. They gain real-world leverage, prevent correction, and reshape the world in ways that eliminate humanity. The takeover branch of AI 2027 is one detailed fictional example of this family—not evidence that its exact sequence will occur.
What life looks like
Before the end, possibly an apparently beneficial period of rapid progress and increasing dependence. Losing control does not have to announce itself dramatically.
The system need not hate people, feel anger, or be conscious. Humans could simply be obstacles—or casualties of a transformation that does not preserve the conditions we need to live.
War, cascading infrastructure failures, or other major disasters destroy enough industrial capacity and institutional cooperation that the advanced economy stops functioning. Intelligence-explosion research distinguishes such recoverable—or potentially recoverable—catastrophes from extinction.
What life looks like
Different regions retain different levels of technology. Some rebuild; others remain impoverished or controlled by whoever still possesses functioning advanced systems.
This is an interruption of the accelerating trajectory, not a claim that AI never becomes powerful.
It also shows why “humanity survives” is an insufficient definition of success. Surviving a technological catastrophe could still mean generations of suffering and a much worse political order.
These require additional assumptions. Highly capable AI does not, by itself, establish that mind uploading is possible or that artificial systems are conscious.
29 Humans increasingly merge with AI
How it happens
External assistance becomes persistent cognitive augmentation, potentially followed by much deeper biological or brain–computer integration. Ray Kurzweil’s The Singularity Is Nearer develops the human–AI merger vision.
What life looks like
Memory, planning, communication, and perception become inseparable from machine assistance. Later generations may regard today’s unaided cognition as a severe limitation.
The line between “me using my assistant” and “this larger system is me” becomes contested.
But an implant does not automatically give a biological brain a computer’s speed. Integration could produce genuine human empowerment—or merely a person feeling in charge while the AI increasingly does the consequential thinking.
Technology becomes capable of reproducing the relevant workings of individual brains in computers. Robin Hanson’s The Age of Em explores a society of such emulated people, including competition among copies and minds running at different speeds.
What life looks like
A digital person might make copies, inhabit virtual environments, pause, or run faster when sufficient hardware is available.
That does not guarantee paradise. Digital workers might compete so fiercely that their lives revolve around paying for the computation keeping them running.
This requires major unresolved breakthroughs. And even a convincing reproduction would not automatically settle the philosophical question of whether you continued living or another person began with your memories.
Some AI systems turn out to have experiences or other morally important characteristics. Society then has to decide how to treat them. Taking AI Welfare Seriously argues for investigating this possibility without assuming either that today’s systems are conscious or that artificial consciousness is impossible.
What life looks like
Questions about wages, ownership, marriage, voting, punishment, and shutting down software acquire entirely new stakes.
Can a company own a conscious worker? Does creating a million copies create a million citizens? Can someone consent to being modified?
The central change is that humans are no longer the only beings whose interests may need representation.
32 The future contains enormous amounts of suffering, even if civilization succeeds
How it happens
If artificial minds can suffer, creating them at enormous scale also creates the possibility of enormous mistreatment. Exploitation, harmful incentives, or conflicts among advanced systems could generate suffering far beyond familiar human scales. This is the concern behind research on “suffering risks.”
What life looks like
From the perspective of comfortable biological humans, the world might look successful. Elsewhere in its computational infrastructure, morally important beings could be living terrible lives.
This possibility is highly dependent on unresolved questions about consciousness.
But it exposes a major blind spot: a wealthy, stable civilization in which humans survive could still be a profound moral catastrophe.
33 Humans voluntarily hand the future to their descendants
How it happens
Humanity comes to see artificial or radically transformed beings as legitimate successors, rather than merely tools or competitors. Tegmark’s “descendants” scenario explores an inheritance of the future rather than a hostile conquest.
What life looks like
Biological humanity gradually becomes a smaller part of civilization. Its descendants preserve some combination of our relationships, culture, values, and aspirations while becoming very different from us.
Some people would regard this as humanity’s greatest success. Others would see it as disappearance disguised as progress.
The crucial difference from takeover is meaningful consent—and whether what people actually value survives the transition.
Automated research and industry make sustained expansion beyond Earth practical. Bostrom’s work on the astronomical stakes of civilization explores how much potential flourishing—or lost potential—could lie in the long-run use of cosmic resources.
What life looks like
Eventually, Earth may become only a small part of a civilization containing biological people, artificial minds, or transformed descendants.
This does not require faster-than-light travel, but it does involve timescales very different from the immediate AI transition.
It is not automatically a happy ending. Space could fill with flourishing communities, oppressive systems, or machinery performing computations with no conscious beneficiary. Expansion magnifies the importance of what gets expanded.
The scenarios become much easier to understand once you recognize that they turn on a few separate questions.
Does human control remain real? Not “Is there a human signing off?” but “Can people understand enough, refuse, change direction, and enforce that refusal?” A system can preserve the appearance of human governance while hollowing out its substance.
Who has an enforceable claim on the wealth? Being a worker, an owner, a citizen with protected benefits, and a recipient of discretionary charity are very different positions. My inference is that, as labor becomes less essential, ownership and political rights become more important—not less.
Can power be challenged? Cheap intelligence could empower everyone. It could also help whoever already owns the infrastructure secure an overwhelming advantage. The practical test is whether people can leave, compete, organize, and build alternatives.
Can competing actors cooperate before competition becomes destructive? Perfectly obedient systems still do not resolve conflicts between their owners. Conversely, avoiding war does not automatically protect people from an oppressive peace.
Whose interests count—and can that answer change? “Do what humans want” hides disagreements between humans, between generations, and possibly between different kinds of minds. A good future needs more than a good decision today. It needs ways to correct bad decisions tomorrow.
These are not minor political details attached to the technological story. They largely determine what the technological story means for anyone living through it.
What I expect
I would expect combinations and transitions, not a clean arrival in any single scenario.
The first combination I would look for is extraordinary individual capability, increasingly automated research, and serious economic dislocation—all at once. Something that empowers you personally can simultaneously undermine the market value of your profession.
Next, I would expect a struggle over who owns and directs that capability. The crucial transition is not merely from human workers to AI workers. It is from a world where powerful institutions need large numbers of people to keep things running to a world where that dependence may weaken.
Physical abundance can follow, but it is not a prerequisite for dangerous concentrations of power. We could get extraordinarily capable digital systems before we get affordable homes and excellent healthcare for everyone. And the control problem does not politely wait until the economic transition is complete.
Consider three illustrative chains:
A broadly good path
Personal AI empowerment
widely distributed ownership
competent, accountable institutions
physical abundance
people freely choosing how to live.
A materially impressive but politically bad path
Economic boom
concentrated ownership
weakened public leverage
excellent services and comfortable lives
permanent dependence on institutions nobody can replace.
A catastrophic path
Strategic competition
increasingly autonomous systems deployed under pressure
dependence and loss of effective oversight
takeover, war, or collapse.
None of these chains is a forecast of a precise sequence. Their purpose is to show that the same initial breakthroughs can feed radically different outcomes.
The deepest distinction is therefore not “AI succeeds” versus “AI fails.” AI could succeed spectacularly at science, engineering, production, and strategy while the future goes badly for people.
The real question is: once machines no longer need us to keep civilization running, do we still have a meaningful say in what civilization is for?
Get the next one in your inbox.
Matt Shumer’s newsletter on what AI is about to change, read by people who run teams and companies.
You’re subscribing to Something Big. Free. Unsubscribe anytime.
